Comment by newyankee
there is enough proof that they had a chatbot internally which was quite competitive but was not pushed through for all these fears, it seems they were always confident that they could catch up and scaling laws were their internal defense.
The question now though is neither might have expected Chinese labs to catch up so fast.
China releasing open models only helps the big companies make more efficient inference.
Maybe they don’t realize that the money will be in the inference compute and there is limited applicability for low flops inference.
Ie. All the breakthroughs they share for free will immediately improve profitability of the ai compute clusters.
Not sure why people think otherwise.